[parisc-linux] State Of The Port

[parisc-linux] State Of The Port

[Matthew Wilcox]

Summary of the telephone meeting held 7/Jan/2000.

Participants:

Michael Ang
Paul Bame
Grant Grundler
R. Scott Holbrooke
David Kennedy
John Marvin
Martin Peterson
Philipp Rumpf
Alex deVries
Matthew Wilcox

1. The Linux 2.3 tree will be opened up today for the general public.

2. The PCI code compiles but has not yet been tested.  Extra work will
   need to be done to support PAT PDC, and Grant will look into this.

3. David will get the Tulip driver working.

4. Martin volunteered to take care of the serial driver problems we are
   experiencing

5. John & Philipp will work on the TLB miss handlers which are currently
   limiting development in a couple of areas.

6. Matthew will work on the proper syscall code.

7. Philipp has the fork() code working, and will next examine exec().

8. Scott volunteered to take care of writing TCP and UDP checksumming
   routines in PA-RISC assembler.

9. Matthew will arrange for a suitable `ifconfig' binary to be available
   for the purposes of testing the network driver.

10. Michael will work on implementing more HPUX syscalls in Linux to
    allow Apache to work.

11. David volunteered to set up a random config test system.

12. Matthew will remove some config options which are known to not work
    (and are unlikely to work in the near future).

13. Grant is currently building a 712 system for Sammy who will be
    transferring his energies from the sun3 port.

14. Frank Rowand is leaving HP.  He has a couple of loan machines
    and may continue to contribute to the port if he has spare time.
    Paul volunteered to liase with Frank to look at the kernel debugger
    code that Frank had written.

15. We discussed a possible hackfest in Fort Collins, and agreed to it
    in principle, but not before Feb 1st.


P.S.  I was tempted to start this email with `The Linux/PA-RISC Core
Team had a Secret Meeting and have deigned to share the results with the
Unwashed Masses', but given how much trouble my last sarcastic comment
caused, I decided to refrain.

[parisc-linux] State Of The Port

[Michael Ang]

Here are Grant's notes from today's meeting with a few annotations:
---
Attendees (in no particular order):
        Mike Ang
        Thomas Bogendorf
        Mathew Wilcox
        Grant Grundler
        Martin Petersen
	plus some lurkers

Tulip/Networking We're still having a problem with media selection and
interrupts on A180, and endianness problems on 21140 chipset PCI cards.
Fleedwood made good progress with a pcnet32 card with endian clean driver,
exposing bugs in the tcp checksum code which have been largely fixed.

Thomas Bogendorf will be getting tulip working very soon.
    Expects to have it working by this weekend. Lack of endianess
    support is the main problem.  2.2.14 is the "latest" tulip.c
    but has no big endian support.

Mike Ang will send the outputs of the tulip diagnostic program in
cvsroot:/userspace/ktulip-diag.c to Donald Becker for help.


SuperIO serial

SuperIO serial support is required to get userland serial console working
on C3000.

Martin Petersen looked at data sheet and it's way too complicated.
Grant Grundler: I think the various components have to share IRQ lines.
        Since this bugger is 3 PCI functions and only function 1 (or 2)
        has the IRQ line, it gets a bit messy.

        I think we also need to get some specifics on how firmware
        *reports* those devices. I helped set some of that up for HP-UX
        but that was over a year ago.
        IIRC, the function 0 gets advertised first to HP-UX but it's
        interrupt comes through function 1. So the driver for function 0
        has to poke registers on function 1.

Grant - will check if anyone in HP is interested in making this work.
Mike volunteered to start looking at it on Monday (2/7/2000).


Interrupts on C3000
Grant Still working on it.
Fleed is still having/had trouble with card-mode dino.
    dino.c.ilr data page faulted. Grant will look at this
    after spending some more time on C3k problems.

Willy mentioned that we will have to think about running on 64bit at some
point.


Short term goals (roughly in order)
- serial console input and output
- networking
- nfs root filesystem
- scsi support

Re: [parisc-linux] State Of The Port

[Thomas Bogendoerfer]

On Fri, Feb 04, 2000 at 11:17:33PM -0500, Michael Ang wrote:
>         Thomas Bogendorf

look it up in the From: line:-)

>     Expects to have it working by this weekend. Lack of endianess
>     support is the main problem.  2.2.14 is the "latest" tulip.c
>     but has no big endian support.

the 2.2.14 tulip.c driver has big endian support as well as Donald's latest 
alpha version.

> Willy mentioned that we will have to think about running on 64bit at some
> point.

IMHO, it's way too early to even think about it. Of course 64bit is sexy,
but without a working 32bit userland it's a too big can of worms to open
right now. Linux/MIPS64 hasn't started before a reasonable distribution
existed, and I'm sure this wasn't a bad descision then.

Thomas.

-- 
   This device has completely bogus header. Compaq scores again :-|
It's a host bridge, but it should be called ghost bridge instead ;^)
                                        [Martin `MJ' Mares on linux-kernel]

[parisc-linux] ELFness and 64bit

[willy]

On Sat, Feb 05, 2000 at 03:37:59PM +0100, Thomas Bogendoerfer wrote:
> > Willy mentioned that we will have to think about running on 64bit at some
> > point.
> 
> IMHO, it's way too early to even think about it. Of course 64bit is sexy,
> but without a working 32bit userland it's a too big can of worms to open
> right now. Linux/MIPS64 hasn't started before a reasonable distribution
> existed, and I'm sure this wasn't a bad descision then.

We want to get people working on userland as soon as possible since this
parallelises well with other people hacking on the kernel.  However, we
need to go 64-bit inorder to support the L-class and similar machines.
I also mentioned that we need ELF support RSN as it is hurting us quite
badly.  I know Sammy has been working on getting 32-bit ELF going on
his 712 for the kernel.  Sammy, how's that going?

Re: [parisc-linux] State Of The Port

[Grant Grundler]

Michael Ang wrote:
> 
> Here are Grant's notes from today's meeting with a few annotations:

Mike,
thanks for touching up the notes...apologies to Thomas for mangling
his name. My error.

> 
> Interrupts on C3000
> Grant Still working on it.

I think it's working now. It still needs more work but folks could
start using it at least. I expect the code to work on j5k but
don't know if either processor has "0xfffa0000" HPA. If one does,
it should work too. All I/O interrupts will be directed to that
processor for now.

Feel free to remove the:
	do_irq_mask 00200000 c02d1324 c01f6500
	iosapic_interrupt: irq 266 line 1 eoi fed32840

messages if they get in your way.

I tested using eepro100.c driver (CONFIG_EEXPRESS_PRO100=y)
and added "#define USE_IO" near the top. I can't say the interface
is totally functional (I haven't the setup to test ping) but the hub
seems happy.


> Fleed is still having/had trouble with card-mode dino.
>     dino.c.ilr data page faulted. Grant will look at this
>     after spending some more time on C3k problems.

I'm going to look at this next before working more on iosapic.c.
Thomas is going to send me console output and I'll see what I
can make out of that.

> Short term goals (roughly in order)
> - serial console input and output
> - networking
> - nfs root filesystem
> - scsi support

More important than any of these, IMHO, is FDB. And I forgot to
mention it at the meeting. Future progress would be substantialy
faster if we had a debugger. I don't feel competent to work on
that. Anyone interested in (or already) pursuing this?

thanks,
grant

Grant Grundler
Unix Development Lab
+1.408.447.7253

Re: ELFness and 64bit

[Thomas Bogendoerfer]

On Sat, Feb 05, 2000 at 01:26:13PM -0500, willy@thepuffingroup.com wrote:
> We want to get people working on userland as soon as possible since this
> parallelises well with other people hacking on the kernel.  However, we
> need to go 64-bit inorder to support the L-class and similar machines.

sure, but there aren't enough people to split between even more machines.
I'd say let one person look into 64bit issues for now, and do the real
work, when 32bit is (more) solid.

Thomas.

-- 
   This device has completely bogus header. Compaq scores again :-|
It's a host bridge, but it should be called ghost bridge instead ;^)
                                        [Martin `MJ' Mares on linux-kernel]

Re: [parisc-linux] ELFness and 64bit

[Sam Creasey]

On Sat, 5 Feb 2000 willy@thepuffingroup.com wrote:

> On Sat, Feb 05, 2000 at 03:37:59PM +0100, Thomas Bogendoerfer wrote:
> > > Willy mentioned that we will have to think about running on 64bit at some
> > > point.
> > 
> > IMHO, it's way too early to even think about it. Of course 64bit is sexy,
> > but without a working 32bit userland it's a too big can of worms to open
> > right now. Linux/MIPS64 hasn't started before a reasonable distribution
> > existed, and I'm sure this wasn't a bad descision then.
> 
> We want to get people working on userland as soon as possible since this
> parallelises well with other people hacking on the kernel.  However, we
> need to go 64-bit inorder to support the L-class and similar machines.
> I also mentioned that we need ELF support RSN as it is hurting us quite
> badly.  I know Sammy has been working on getting 32-bit ELF going on
> his 712 for the kernel.  Sammy, how's that going?

heh...  welll....   I'm getting a better understanding of parts of the
problem, if that makes anyone feel better. :)

I've actually got a kernel together than compiles and links in a fairly
sane manner...  except for the .data segment, which will, eventually, get
badly munged..  It's looking like a linker error revolving around certain
types of relocations that don't quite work right.  further investigation
is still needed, but I've not had time for the last week or so.